jeyoor/pymich-dl

Deep Learning Talk for PyMichiana Meetup

Outline for 2019-10-02 PyMichana Talk

Followup:

• ComputerVision Renaissance

  • https://www.embedded-vision.com/technology/computer-vision-algorithms
  • https://www.udacity.com/course/introduction-to-computer-vision--ud810

• Spotify Genres

  • https://benanne.github.io/2014/08/05/spotify-cnns.html
  • https://papers.nips.cc/paper/5004-deep-content-based-music-recommendation

• Google Doodle: Music style transfer (Bach style)

  • https://www.google.com/doodles/celebrating-johann-sebastian-bach
  • https://magenta.tensorflow.org/

• Generative Adversarial Networks

  • https://medium.com/sigmoid/a-brief-introduction-to-gans-and-how-to-code-them-2620ee465c30
  • https://towardsdatascience.com/gan-by-example-using-keras-on-tensorflow-backend-1a6d515a60d0
  • https://machinelearningmastery.com/how-to-train-stable-generative-adversarial-networks/

• Activation Functions Explanation

  • https://missinglink.ai/guides/neural-network-concepts/7-types-neural-network-activation-functions-right/
  • http://robotics.stanford.edu/~amaas/papers/relu_hybrid_icml2013_final.pdf
  • https://arxiv.org/pdf/1710.05941v1.pdf

• Information Theory for Deep Learning (startings of a theoretical framework)

  • https://www.youtube.com/watch?v=XL07WEc2TRI
  • https://lilianweng.github.io/lil-log/2017/09/28/anatomize-deep-learning-with-information-theory.html
  • https://www.quantamagazine.org/new-theory-cracks-open-the-black-box-of-deep-learning-20170921/
  • http://naftali-tishby.mystrikingly.com/

• WaveNet

  • https://cloud.google.com/text-to-speech/docs/wavenet
  • https://deepmind.com/blog/article/wavenet-generative-model-raw-audio

• PyTorch and Keras Comparison

  • https://deepsense.ai/keras-vs-pytorch-avp-transfer-learning/

• 3D Training Environments

  • https://gym.openai.com/
  • https://gym.openai.com/envs/Humanoid-v2
  • https://github.com/deepmind/lab

• Using Deep Learning for Text & Language (Text Transformers)

  • https://openai.com/blog/better-language-models/
  • https://jalammar.github.io/illustrated-gpt2/

• cnn visualization

  • machine learning: how do we use statistics (and software) to make a simple decision (simple output) based on a complex situation (complex input)
    • detecting music genres (audio classification)
    • placing images into categories (image classification)
      • detecting cancer in body scans
    • self-driving cars (reinforcement learning)
  • learning uses a lot of data
    • each eye is transmitting around 10 megabits per second
    • the imagenet dataset has over 14 million images
    • typical "comfortable" use for a private dataset would be about 100,000 images
    • you need lots of data to make your machine learning project successful
  • deep learning: machine learning using an artifical neural network (ann) with many layers (usually processed on a GPU for massive paralellism)

• dogs/cats python notebook

  • talk about collaboratory
  • show the Keras code setting up the neural network
  • talk about using sklearn-metrics and confusion matrices for evaluation

• More details on Convolutional Neural Networks

  • what's a convolution?

WARNING: This has been the briefest overview. You need more depth to really grok this stuff.

See further:

• Standford Coursera Course on Machine Learning
• 3 Blue 1 Brown Youtube Channel
• fast.ai

Contact Info

jey@mantleup.com github.com/jeyoor/pymich-dl

Possible Future Meetup Topics

• pygame
• ChiPy or Indi Py
• graphic novel/interactive fiction in python
• python for IoT (micropython)